Chaotic multi-strategy adaptive walrus optimizer for global optimization and feature selection

The walrus optimizer (WO) is a new meta-heuristic algorithm designed to mimic the behaviors of oceanic walruses. However, the consistency of the results is impacted by modifications to the WO parameters. Eventually, the algorithm becomes constrained to local solutions, because minimizing the boundary is likely to lead to overlapping solutions. To address WO’s shortcomings, the proposed version of WO introduces the multi-strategy walrus optimizer (MSWO), an integrated version of WO with several purposeful strategies. Chaotic operator strategy, adaptive strategy, mutation strategy that introduces slight random changes to the search agents, and dynamic boundary location micro-adjustment strategy are some of these purposed strategies. First, a promising chaotic map that expands the search space and provides diversity is used as part of a chaotic approach to stop WO from converging in the early phases of optimization. Second, instability caused by changes in WO parameters is addressed using an adaptive technique. To enhance the population variety of the basic WO, a linear scaling approach is then applied to modify the search agents’ positions inside the dynamic border. The purpose of the dynamic learning procedure is to improve the ability of search agents to adapt to changing locations in the search space. Finally, to improve the speed of MSWO and help it break out of local optimums, a Gaussian mutation strategy is included, which determines the size and direction of these changes using a Gaussian distribution. In evaluation studies, the performance of the proposed MSWO algorithm was confirmed using the CEC2019 and CEC2020 benchmark test functions. The proposed MSWO algorithm was then used to address five conventional engineering problems to demonstrate its dependability and suitability for real-world problems. Furthermore, a few feature selection (FS) problems were used to assess MSWO’s applicability to challenging real-world optimization problems. The results of the evaluation experiments show that the proposed MSWO algorithm performs better than the basic WO method and other well-known algorithms.